嗨，我是 Dustin。
Hi, I'm Dustin.
教了 12 年 GMAT，現在人在曼谷，用 AI 做教育產品。這裡寫的是一個沒有工程背景的人，怎麼靠 Claude Code 和 Gemini 把東西做出來的過程，還有途中踩過的坑。
12 years teaching GMAT. Now based in Bangkok, building education products with AI. This blog is about how someone with zero engineering background actually ships things — and all the pitfalls along the way.
精選
-
The More Rules You Add, the Less Claude Listens — I Sent a Team of Agents to Trim My Setup and Cut 36% of Always-On Context
Late at night I got Opus 4.8, and my weekly quota happened to reset. The first thing I did was put my harness on a diet. A COO student who had burned through 88% of his 1M context made me face one thing squarely: the more rules you add, the less the model listens.
-
規則越加，Claude 越不聽話——派一隊 AI 重整設定，常態上下文省 36%
半夜拿到 Opus 4.8、額度剛好重置，我做的第一件事是減肥我的 harness。一個 COO 學員把 1M 上下文用到 88% 的案例，讓我認真面對一件事：規則加越多，模型反而越不聽話。
-
The Truth and BS of Vibe Coding: Observations from a 10-Year Teaching Veteran
We shouldn't uniformly trash Vibe Coding, but we do need to recognize the traps in course quality. A teacher with 10+ years in the game shares how to tell the real deal from the fakes.
-
Vibe Coding 的真與假：一個從業十年老師的觀察
不該統一貶低 Vibe Coding，但也要認清課程品質的陷阱。一個教學從業十年以上的人，分享辨識有料與沒料的方法。
-
The Dunning-Kruger Curve of Claude Code: From Clueless to Perfect Sync
A heavy user with zero engineering background shares the Dunning-Kruger curve of Claude Code from day one to week three, and the workflow that finally clicked.
-
Claude Code 的達克曲線：從陌生到極致同步的實戰心得
一個非工程師背景的重度使用者，分享 Claude Code 從第一天到第三週的達克曲線效應，以及最終找到的高效互動方式。
最近在寫
-
ADHD, Flow, and Why I Teach AI
-
AI Micro-Notes 2026: Thoughts Too Short to Trash
-
The Client Says They Want Training, But Deep Down They Want a System
-
From Takedown to Return, I Caught a Case of AnthroPTSD
-
Claude Seeing Ghosts Four Nights Straight: A Log of Opus 4.8 Fabricating Tool Output
-
ADHD、心流，與我為什麼教 AI
-
AI 碎念日記 2026：那些太短但捨不得丟的觀點
-
客戶嘴上要培訓，骨子裡要系統
-
從下架到回歸，我得了 AnthroPTSD
-
Claude 半夜見鬼連續四天：Opus 4.8 捏造工具輸出實錄
-
Opus 4.8 Cries 'Prompt Injection,' Codex GPT-5.5 Tracks Down the Real Cause: a Worktree Race
-
Opus 4.8 搞烏龍報『提示注入攻擊』，Codex GPT-5.5 揪出 worktree race 真因
-
Running Taiwan Sports-Lottery Odds With Mobile Claude Code, Standing in the Lottery Shop: the Odds Are Garbage, You'll Lose
-
在彩券行用手機 Claude Code 算運彩賠率：台運賠率爛透了，誰下誰賠
-
Which AI Is Strongest? I'll Take Them All: Claude as the Brain, Directing Gemini and Codex
-
哪個 AI 最強？我全都要：讓 Claude 當大腦，指揮 Gemini 與 Codex
-
No Error Doesn't Mean Success: Five Silent Failure Traps in AI Dev
-
沒拋錯不代表成功：AI 開發裡的五個靜默失敗陷阱
-
Two Mindsets for Getting Unstuck: Logical Thinking, and the AI Pedagogy of Never Making the Same Mistake Twice
-
卡關時的兩個心態：用邏輯思考，與『不貳過』的 AI 教育學
-
Teaching a Talent Agency to Use Claude Code, and the Soft Landing of the DK Confidence Curve
-
帶藝人經紀公司用 Claude Code，與 DK 信心曲線的軟著陸
-
My First All-English Two-Hour Corporate Training: I Burned the Script Two Hours Before
-
第一次全英文兩小時企業內訓：上課前兩小時燒掉提詞稿
-
The Perpetual Token-Burning Machine: Anthropic Really Is Trying
-
燒 token 永動機：Anthropic 真的很努力
-
The Five-Step SOP for Sneaking Claude Code Runs From the Hotel Over Dinner
-
出門吃飯前，讓 Claude Code 在旅館背景偷跑的五步 SOP
-
From Miasma to Hades: How One Group Turned AI Tools Into a Supply-Chain Attack Vector
-
從 Miasma 到 Hades：同一個攻擊組織如何把 AI 工具當成供應鏈攻擊媒介
-
Reading the Claude Fable 5 / Mythos 5 System Card Feels Like a Sci-Fi Novel
-
Testing Fable's Limits on a Professor Friend's Paper Repo
-
翻 Claude Fable 5／Mythos 5 系統卡，越看越像在讀科幻小說
-
拿教授朋友的論文 repo 試 Fable 的能力邊界
-
A Few Takeaways from the Claude Code Anniversary Talk
-
Don't Dump the Last Session's JSON — Check Your Harness First
-
Claude Code 週年對話的幾個重點
-
別把上一個對話的 JSON 餵給它，先體檢 harness
-
What to Know Before Handling Sensitive Data with Claude
-
用 Claude 處理敏感資料前必知
-
Worried About Agents Going Rogue? Start Isolated, Hand Off Gradually
-
A Few Months of AI Teaching Content: What Gets Traffic and What Doesn't
-
"Automation" Isn't the Answer to Most Things — and How I Made This Video
-
A Few Months With Claude Code: The Companion Tools and Resources I've Accumulated
-
Claude Code To-Do Tiering: Auto / Review / Collaborate / Manual
-
Digital Nomadism Is Something You Stumble Into, Not a Goal to Chase
-
The Teleprompter That Saved My English-Only Teaching, and the Open-Source AI Meeting Copilot I'm Evaluating
-
A Week of Firestore / Cloud Functions Footguns — Same GMAT Question Bank
-
Three Real Lessons From Going Freelance, Plus the Pros Who Hide in Plain Sight
-
Eight Days of Silent Failure: launchd Fired on Schedule, Nothing Happened
-
Model Picks Beyond the Main One: Notes I Left at Different Times
-
Letting AI Write CAD From an Engineering Drawing: A Single Vision Model Can't Be Trusted to Read Topology
-
The Day Opus 4.8 Went Haywire: tool call cannot be parsed
-
A Week of Making Tutorial Videos With Claude Code: Remotion Animation and ffmpeg Pitfalls
-
A Proposal Is Not a Deployment: Why One Quick Win Sat Dead for Five Weeks
-
I Built a "Tinghao SKILL" — Distilling a Finance YouTuber's Perspective Into AI
-
擔心 agent 失控？先在隔離環境逐步放手
-
做 AI 教學內容幾個月：什麼有流量、什麼沒有
-
「自動化」不是大部分事情的解方——以及我怎麼拍這部影片
-
用 Claude Code 這幾個月，累積下來的周邊工具與學習資源
-
Claude Code 待辦分級制：自動／審核／協作／手動四級分流
-
數位遊牧是可遇不可求，不要把它當成追求目標
-
全英文授課的提詞器救星，以及我正在評估的開源 AI Meeting Copilot「Natively」
-
一週 Firestore / Cloud Functions 踩坑合輯——同一個 GMAT 題庫系統
-
出來接案的真實感受三條，還有高手大隱隱於市
-
靜默連敗 8 天：launchd 照常觸發，卻什麼都沒發生
-
主力之外，各模型的定位：我在不同時間留下的選型碎念
-
讓 AI 看工程圖寫 CAD：單一視覺模型裸寫拓樸不可信
-
Opus 4.8 全線抽風的一天：tool call cannot be parsed
-
用 Claude Code 做教學影片一週踩到的坑——Remotion 動畫與 ffmpeg 後製
-
提案不等於部署：一個 Quick Win 掛了五週的根因
-
我做了一個「庭皓 SKILL」——把財經 YouTuber 的觀點蒸餾成 AI
-
The Blind Spot in AI Self-Verification — Why You Need an Uninformed Agent to Audit
-
AI 自我驗證的盲點——為什麼要請一個不知情的 agent 來稽核
-
Better Context Beats a Stronger Model — repowise's Five-Layer Architecture
-
給 LLM 更好的 context，勝過換更強的模型——repowise 的五層架構
-
Two Reminders for Vibe Coders: Learn System Architecture First, the Value Is in the Thinking
-
給 Vibe Coder 的兩個提醒——先學系統架構，價值在思維不在工具
-
Don't Brute-Force Sites That Need a Login—Your Token Is Your Account
-
需要登入的網站別硬爬——你的 token 就是你的帳號
-
A Student's Seven-Agent Setup That Wouldn't Run — A Cautionary Tale About Harness First
-
學員自建七個 agent 跑不動——一個 harness 為重的反面教材
-
Leveling Up the iPad Workflow — Surviving Internet and Power Outages, and Why I Ditched the Magic Keyboard
-
iPad 工作流進階——斷網斷電的備案，與我為什麼放棄妙控鍵盤
-
Love-Hate With Anthropic — Sneering at the Elitism While Devouring Their Engineering Blog
-
又愛又恨 Anthropic——一邊嫌精英心態，一邊狂讀他們的技術部落格
-
Racing for the Fastest Summary — Opus 4.8's 244-Page System Card, and How I Read It With 20 Agents in Half an Hour
-
拼全網最速——Opus 4.8 系統卡 244 頁重點，外加我怎麼用 20 個 agent 半小時讀完
-
AI Tool Philosophy — Pick a Mindset, Not a Side
-
AI 工具使用哲學——不選陣營，選思維
-
Claude Code Ecosystem Overview — Skills, Hooks, Agents in Full Bloom
-
Claude Code 生態系盤點——Skill、Hook、Agent 全面開花
-
Turning Street Smarts into an AI Skill — When the Numbers Are Clear, You Stop Flinching
-
把《街頭智慧》做成 AI Skill——算出來的數字就在那邊，波瀾不驚
-
Every AI Leader Starts Cutting Corners — The GPT, Claude, Gemini Cycle
-
坐上第一就開始拿翹——GPT、Claude、Gemini 的降智循環
-
Running Claude Code on iPad — tmux + Tailscale + Moshi Setup Guide
-
iPad 跑 Claude Code 全攻略——tmux + Tailscale + Moshi
-
Cross-Model Review — Stop Letting AI Grade Its Own Homework
-
跨模型互審——讓 AI 不再自己審自己
-
Stop Dismissing Gemini — Four Use Cases Where Nothing Else Comes Close
-
別一味貶低 Gemini——四個其他家打不過的 Use Case
-
Gemini 3.5 Flash Reddit Reviews — 3x Price, Vision Regression, Tool Calling Disaster
-
Gemini 3.5 Flash Reddit 實測彙整——貴三倍、Vision 退步、Tool Calling 災難
-
Have Claude scan an MCP before you install it—it found 7 vulnerabilities and still said it's safe to install
-
裝 MCP 之前先讓 Claude 幫你掃——找到 7 個漏洞，但說這個可以裝
-
Scope Discipline: five rules—client small talk ≠ scope expansion authorization, must go into every future contract
-
Scope Discipline 五條——客戶閒談 ≠ scope 擴張授權，未來合約必含
-
YouTube Data API large-file upload tested: 216 MB succeeds, 257 MB hangs—threshold around 220-250 MB
-
YouTube API 大檔上傳卡死實測——216 MB 過 / 257 MB 不過，門檻約在 220-250 MB
-
MBA × AI Case Method—avoiding HBS licensing risk, finding open case materials, training seed instructors
-
MBA × AI Case Method——避開 HBS 授權風險、找開源案例素材、訓練種子教師
-
Three moves to firepower-demo Agentic AI at your company—desktop cleanup, conflicting meeting notes, multi-version emails
-
在公司火力展示 Agentic AI 的三招——桌面整理、會議紀錄、信件多版本
-
Reflection after a 200-person online AI lecture—the energy of the classroom 10 years ago, back again
-
200 人線上 AI 講座結束後的感言——10 年前實體教室那種時光，又回來了
-
Why Obsidian is something special to me, and why I only made this video now
-
Obsidian 為什麼是個特別的存在——我為什麼到現在才拍這部影片
-
Seven gotchas from running Resend and Gmail email workflows—HTML templates, scheduled_at, Cloudflare, open-rate alignment
-
Resend 跟 Gmail 寄信工作流踩坑七連——HTML 模板、scheduled_at、Cloudflare、開信率對齊
-
Covey's time management matrix, now with Claude Code—how each quadrant shifts
-
Covey 七個習慣的時間矩陣放上 Claude Code 之後——每一格的時間分配怎麼變
-
Two MCP servers for Taiwan public data—what mcp-taiwan-legal-db and Twinkle Hub are each good for
-
台灣公開資料兩個 MCP 評測——mcp-taiwan-legal-db 跟 Twinkle Hub 各適合什麼場景
-
What I only learned after taking on enterprise AI training—prep time, the client liaison, instructor flexibility, and how to pick a consultant
-
親自接 AI 企業培訓之後才知道的事——準備時間、甲方對接、講師應變、選顧問訊號
-
How to tune Opus 4.7's effort, and why I still keep non-coding tasks on Sonnet 4.6 medium
-
Opus 4.7 的 effort 怎麼調，跟為什麼我非編程任務還是停在 Sonnet 4.6 medium
-
Claude Code Desktop vs CLI right now—why I think beginners should just start with the desktop app
-
Claude Code 桌面版跟終端機現在差在哪——我為什麼覺得初學者直接從桌面版開始就好
-
2026 Model Personality Watch: Gemini, Claude, Codex Compared
-
2026 模型脾氣觀察：Gemini、Claude、Codex 的個性對比
-
Why You Keep Hitting Limit: Six Observations on Claude's Subscription Economics
-
為什麼你會一直 hit limit：Claude 訂閱經濟學的六個觀察
-
Three Channels for Tracking Anthropic
-
追蹤 Anthropic 動態的三條管道
-
Jack Dorsey Cut 40% and Rebuilt: The Shape of Companies in the AI Era
-
Jack Dorsey 把公司裁掉 40% 重建：AI 時代的組織形狀
-
Judging AI Risk on Two Axes: Reversibility × Environment Isolation
-
判斷 AI 風險的兩個維度：可逆性 × 環境隔離
-
GPT-image-2 Wants Fewer Constraints — Plus a Consistency Drill
-
GPT-image-2 越少約束越好，加上一個一致性練習
-
Moving from Claude Code to Codex: I Built a Skill to Handle It
-
從 Claude Code 搬家到 Codex：我做了一個 Skill 自動處理
-
You're mass-applying with AI. They're mass-screening with AI. Who wins?
-
你用 AI 海量投履歷，他用 AI 海量篩履歷：最後誰贏了？
-
Is your AI subscription worth it? A night market steak framework
-
AI 訂閱費值不值？夜市牛排思考框架
-
Lost 10 kg After Ten Years of Failure — Because of Claude's Quota
-
Teaching non-technical people AI: five things the tech world assumes everyone knows
-
睽違十年瘦了 10 公斤——竟然是因為 Claude 額度不夠
-
帶麻瓜學 AI：技術圈習以為常、但他們從來不知道的五件事
-
AI community digest W18: Claude detects Codex cheating, GPT 5.5 guardrails confuse everyone
-
AI 社群週報 W18：Claude 偵測到你在偷用 Codex、GPT 5.5 的過濾器到底鬆了沒
-
1M context isn't about holding more — it's about managing better
-
Knowledge base deep clean: using AI to find months-old problems
-
1M Context 不是裝更多，而是管理更好：Claude Code Session Management 心法
-
Claude Code this week: 1M context as a management problem, quota anxiety goes collective, SpaceX takes Cursor
-
Two Worlds of Web-Based Claude Code Setup
-
See the Outside, Think the Inside: Two Claude Skills That Pair
-
Three design mistakes I keep seeing in enterprise AI training
-
Nine video production bugs I hit in April 2026
-
2205% subscriber growth in two weeks: how one video opened the algorithm
-
Claude Code 這週整理：1M Context 怎麼用、額度焦慮集體發作、Cursor 被 SpaceX 搶走
-
網頁版 Claude Code 環境設定的兩個世界
-
看清外面、想清裡面：兩個 Claude Skill 的分工
-
幫企業導入 AI，教案設計最常犯的三個錯
-
知識庫大掃除：用 AI 幫你找到埋了幾個月的爛帳
-
自動化影片製作管道：2026 年四月踩過的九個坑
-
頻道訂閱兩週漲 2205%：一支影片怎麼把演算法打開的
-
A Legal Counsel Used Claude Code to Catch Patent Infringement — Days Down to Hours
-
法務用 Claude Code 抓專利侵權，從幾天變成一天
-
My Claude Code Automation Stack: /loop, Monitor Tool, the Skill System, Obsidian Reflection Reviews
-
我家的 Claude Code 自動化工作流：/loop、Monitor Tool、skill 體系、Obsidian 反思週報
-
Six Months with Claude Code: From First Affection to Hesitation, From Overtime to Detox, From Mandarin Circles to the Thai Community
-
Claude Code 用了半年：從好感到猶豫、從加班到 detox、從中文圈到泰國社群
-
Two Months of Claude Code Harness Ops: Crashes, Zombies, Cache Misses, and Plan Mode Edges
-
Claude Code Harness 兩個月運維史：當機、殭屍、快取、Plan Mode 的邊邊角角
-
AI-Era Business Routes: Auto-Generated Posts, the Skill Poisoning Theory, and Why I Started Rooting for OpenAI
-
AI 時代的經營路線之爭：自動化貼文、Skill 投毒論、跟我期待 OpenAI 的理由
-
A Teacher's Responsibility: How a 15-Year Cram School Instructor Sees AI-Era Course Quality
-
教學者的責任心：15 年補教老師怎麼看 AI 時代的課程品質
-
Claude Code This Week: Offloading CLAUDE.md, Quota-Saving Tricks, Ghostty Click Fix
-
Claude Code 這週整理：分擔 CLAUDE.md 的進階設定、省 quota 技巧、Ghostty 點擊設定
-
Opus 4.7 After One Week: From the System Card to the Roasting, Where Did It Go Wrong?
-
Opus 4.7 上線一週：從系統卡到火烤文，到底哪邊出了問題
-
90% of People Are Two Years Behind: The Real AI Adoption Gap
-
Claude Code Subscription Guide: The Real Gap Between Pro and Max
-
Why Claude Code Only Reaches Its Full Potential in the Terminal
-
90% 的人落後兩年：AI 採用的真實落差
-
Claude Code 訂閱怎麼選：從 Pro 到 Max 的真實落差
-
為什麼 Claude Code 在終端機裡才是完全體
-
What Anthropic Did This Time: Cache Reduction and Token Anxiety Injection
-
Claude Desktop App Got a Redesign — Here's Why I'm Still Using the Terminal
-
hermes-CCC: Bringing All 46 Hermes Agent Capabilities into Claude Code
-
Obsidian vs NotebookLM: Which One Works for Claude Agent Automation?
-
Anthropic 這次幹了什麼：快取縮水與 Token 焦慮注射
-
Claude Desktop App 改版了——以及我為什麼還是用終端機
-
hermes-CCC：把 Hermes Agent 的 46 個能力全部裝進 Claude Code
-
Obsidian vs NotebookLM：用 Claude Agent 自動化知識管理，選哪個？
-
This Week in the Claude Code Ecosystem: Four Resources I Bookmarked
-
本週 Claude Code 生態觀察：四個我收藏的新資源
-
Two HTML/PDF Rendering Traps: Chrome Print Margin and qlmanage's Single-Page Problem
-
兩個 HTML / PDF 渲染陷阱：Chrome print margin 和 qlmanage 的 1 張縮圖
-
Digging into Claude Code's JSONL: Manual Context Usage, Cache Types, and Compact Detection
-
挖 Claude Code 的 JSONL：手動算 context 用量、分辨快取類型、偵測 compact
-
cc-audit: A Skill That Forces You to Save Claude Code Quota
-
cc-audit：一個強迫你省 Claude Code 額度的 Skill
-
Make a Claude Code Tutorial Video in 30 Minutes with Remotion Skill
-
用 Remotion Skill 半小時拍一支 Claude Code 教學影片
-
Is Claude Getting Dumber? Someone Finally Brought Data
-
Claude 變笨了？這次有人拿出數據證據
-
Claude Code's Cache Crisis: Why Your Quota Burns So Fast
-
Claude Code Skills Demystified: Lazy Loading and the Real Trigger Rate
-
Claude Code 的快取危機：為什麼你的額度燒得那麼快？
-
Claude Code Skills 機制解密：按需載入與觸發率真相
-
Claude Code Source Leak: Three Secrets Unearthed by Reddit
-
Claude Code 源碼洩漏：Reddit 大神挖出的三個秘密
-
PII Guard TW: A De-identification Tool Built for Taiwan
-
PII Guard TW：為台灣打造的個資去識別化工具
-
Anthropic's Trust Crisis and My Backup Plan
-
Anthropic 信任危機與我的備援方案
-
Mirror System: Weekly Reviews Without the Comfort
-
Mirror System：用 Claude 做不帶安慰的週回顧
-
Why Does Cowork Hit Your Limit in Hours While Claude Code Runs All Day?
-
為什麼 Cowork 幾小時就頂到限額，Claude Code 整天都沒事？
-
The 5 Levels of Claude Code: You Don't Choose to Upgrade, the Ceiling Forces You
-
Claude Code 五個成熟度等級：你不是選擇升級的，是被天花板逼的
-
Turning Books into Claude Skills: From Textbooks to Atomic Habits, the World Is Already On It
-
把書變成 Claude Skill：從教材到原子習慣，國外已經在瘋了
-
The Real Next Level in Claude Code: Context Management and Hooks
-
Claude Code 的真正進階：上下文管理與 Hooks
-
Obsidian + Claude Code: My Daily Knowledge Management Workflow
-
Obsidian + Claude Code：我的日常知識管理工作流
-
One AI Tool to Rule Them All: Why I Only Recommend Claude Code
-
一個 AI 工具打天下：為什麼我只推薦 Claude Code
-
AI Security Boundaries: The Lobster Craze, Presidio, and My Three Rules
-
AI 的安全邊界：龍蝦熱潮、Presidio、與我的三條鐵則
-
Claude Code Workshop Prep Notes: Teaching Non-Coders to Go Terminal-First
-
Claude Code 工作坊備課日誌：帶零基礎學員直上終端機
-
claude-work-timer: Auto-Track Your Claude Code Working Hours
-
claude-work-timer：讓 Claude Code 自動算工時
-
Claude Code Initial Prompt Tips: Six Techniques for Effective AI Collaboration
-
Claude Code 起始指令心法：六個讓 AI 真正聽懂你的技巧
-
What I Learned from 7,843 Claude Code Conversations
-
Claude Code 用了 7843 條對話後的心得
-
Recommended Reading: AI Articles from Early February
-
I scanned 500 Claude Code sessions and the AI stopped making the same mistakes
-
推薦閱讀：二月上旬的 AI 文章精選
-
我掃了 500 個 Claude Code 對話紀錄，然後 AI 就不再犯同樣的錯了
-
Critical Thinking in the AI Era: Playing Dumb, the Feynman Method, and What Future Talent Looks Like
-
AI 時代的思考力：裝傻狂問、費曼學習法，與未來人才的樣子
-
How Teachers Can Embrace AI Without Getting Replaced
-
教學者如何擁抱 AI 而不被取代
-
My AI Workflow: Convergence and Divergence
-
我的 AI 工作流：收斂與發散
-
FIRE Planning with Claude Skills: Talking Retirement with AI
-
用 Claude Skill 做 FIRE 理財規劃：跟 AI 聊退休
-
Building AI Products from Scratch: A Non-Engineer's Four-Generation Evolution
-
從零打造 AI 產品的踩坑日記：一個非工程師的四代進化
-
Can't See Myself in Three Years: Anxiety and Adaptation in the Age of AI
-
看不見三年後的自己：AI 時代的焦慮與適應
-
Claude vs Gemini vs GPT: A Power User's Honest Field Notes
-
Claude vs Gemini vs GPT：一個重度使用者的模型體感實錄
-
Hello World: Why I Started This Blog
-
你好世界：為什麼我開始寫這個部落格
-
AI Micro-Notes 2025: Thoughts Too Short to Trash
-
AI 碎念日記 2025：那些太短但捨不得丟的觀點